Context / Standard term
Semantic search
Search that ranks results using learned representations of query and document similarity, allowing related wording to match without exact terms.
Semantic search commonly embeds the query and indexed text, then ranks nearby vectors. A query for onboarding may retrieve material titled getting started even when the exact word is absent. The model can also group opposites, neighboring topics, or generic text because learned similarity is not the same as answering the question.
Builder example
Semantic retrieval helps when users do not know the collection's vocabulary. Exact identifiers, negation, dates, and policy versions may be better served by keyword or structured filters. Hybrid search and reranking are common options, but the appropriate design comes from labeled queries and access requirements.
A customer searches for 'how do I get a refund' and finds nothing because the help doc is titled 'return and exchange policy.' Keyword search requires the exact word.
Use semantic search so meaning-based queries find the right docs. Combine it with keyword search to also catch exact terms like error codes and product names.
Common confusion: Semantic search finds meaning-similar results, which sometimes means it returns topically related content that does not answer the question. A query for "refund policy" might surface a document about "return shipping" because the meanings are close, even though the content differs.